Obstacle region recognition using single image for planetary soft landing

In order to overcome the problem of obstacle detection in complex terrain for planet soft landing, a passive visual obstacle detection method based on single image was proposed. First, a saliency map of the image was obtained using the saliency detection based on the phase spectrum method, and the variance map of the original image was calculated. Then, the two-dimensional histogram of terrain image was constructed by counting the gray level of two graphs, and the segmented Gamma correction was used to enhance the peak feature. Furthermore the two-dimensional histogram of terrain image was segmented by the two-dimensional Otsu method, which was used to obtain the binary image of the obstacle area. Herein, the various terrain grayscale images provided by HIRISE and the corresponding DEM data were employed to evaluate the accuracy of the obstacle detection algorithm. The average detection rate of TN + TP is over 80%. Experiment result indicates that the algorithm can effectively segment the obstacle area and the safety area in complex terrain environments, thus providing useful information for landing point selection.